Mannerisms

“We’re working with a university in the UK to apply deterministic chaos theory to semiconductor forecasting”, says Malcolm Penn, CEO of Future Horizons.

“Forecasting the semiconductor industry is not an exact science” explains Penn, “the data is not exact. Some companies don’t report data. Sometimes trade bodies make assessments. Any forecast is only accurate to plus or minus 3 per cent. When people change their forecasts by a fraction of a percentage point, forget it, it’s meaningless.”

Penn points out that one rerason why forecasting is difficult is because the semiconductor industry is not maturing, the price wars on DRAM and microprocessors which have waged all year, are evidence of that.

Also making forecasting difficult is the fact that the industry has little control over its pricing. “Very few products are sold on value, the industry is a cost plus business,”, says Penn, “and the cost is $9 per square centimetre. It has been $9 per sq. cm for a long time. It’s been remarkably constant.”

Gordon Moore quantifies that constant cost of silicon using a rather different metric. Moore’s calculation is that processed silicon has always sold for $1bn per acre.

Another factor making semiconductor industry forecasting particularly difficult is the fact that the industry is not homogeneous. The widely different financial results of the companies are evidence of that, with nimble-footed fabless players like Broadcom, out-thinking and out-selling slow-moving IDMs.

“No company is the bellwether for the industry,” says Penn, “there used to be the big broad range companies with some uniformity in their results, but now financial results are all over the place.”

Professed moves to going ‘fab-lite’ are a nonsense says Penn. “You can’t have half a fab, you either have a 32nm fab or you don’t”, says Penn, “saying: ‘We’re going fab-lite at 32nm’ means: ‘We’re going fabless at 32nm’. Yes, companies may still have fabs, but they’ll be analogue fabs or 130nm fabs. But they won’t have advanced CMOS fabs.”

The move by IDMs to fabless will cause huge strains. “Downsizing a semiconductor company is a huge problem. No one’s ever done it successfully yet”, says Penn, “I see a huge change in the next ten years, a lot of top ten companies today won’t be in the 2015 top ten.”

The moves to going fab-lite syndrome reflect the undue influence of accountants. So do the moves to cut back on R&D.

“When the bean-counters take over, it’s all over”, says Penn. He instances cut-backs in R&D. “Cuts in R&D won’t impact your company to three to four years but, after that you’re dead”, he says, “you can cut the R&D and the company’s financial results will look good for three years but, after four years, you’ll be dead. Accountants just don’t understand that.”

TOMORROW MORNING: THE TEN BEST REASONS FOR SEMICONDUCTOR START-UPS